Choose a voice
Ten to pick from, and a button that reads a line in each one.
The numbers are the names: 426, 440, 383, 225 and the rest. They are short, easy to say out loud, and none of them collides with a word already in the menu.
- 1
Open the Voice menu in the menu bar
Ten ship with the app. Two are Supertonic, which needs nothing installed; the other eight are Piper, which uses espeak-ng for pronunciation.
- 2
Press the sample button
It reads a line out loud in that voice. Hearing one is the only way to choose one, and it costs a second.
- 3
Pick it, and it is used from then on
Synthesis runs at about nineteen times realtime, so a five-second reply starts in under a second whichever you choose.
A voice per project
Two checkouts of the same repository get different voices, which means the one asking you a question identifies itself before you have looked at the screen. This is the feature that stops mattering the moment you only have one agent running and starts mattering enormously the moment you have three.
When it goes wrong
The Piper voices are greyed out
Piper needs espeak-ng for pronunciation and it is not on your PATH.
Install espeak-ng, or stay on the two Supertonic voices, which do their own text processing and need nothing. They are the default for exactly this reason.
A voice starts mangling the first word or two
A voice that was downloaded but not completely.
Open Setup. It notices a file shorter than it should be, throws it away and fetches it again rather than trying to load it.
It talks too fast, or too slowly
Speaking pace is a setting, and one pace does not suit everybody.
Set EARSHOT_SPEED. Below one is slower. Anything outside a tenth to three times falls back to normal rather than being clamped.